Workflow disruption cascades in the operating room: an empirical framework for modeling high-risk, high-disruption surgical states

Background The digital revolution in surgery provides new ways to understand accidental harm. Surgical workflow disruptions (FDs) are associated with increased team stress, prolonged procedures, and elevated error rates. We introduce the concept of disruption cascades —sequences of escalating events—classified into three mechanisms: random occurrences (which do not constitute a cascade), common-cause triggers, and concatenating “domino effects.” This study investigates the presence of these mechanisms and evaluates the ability of artificial intelligence (AI) to detect disruption cascades. Methods Flow disruptions from sixteen surgical cases were manually reviewed by a human expert using custom labeling software to identify disruption cascades. ChatGPT-4 was used to assess AI’s ability to detect cascades and measure interrater reliability. Results Among 1,631 FDs, 152 cascades were identified. Approximately two-thirds of disruptions were random (non-cascading). The overall FD rate was 0.43/min, varying by the cascade type in-progress: 0.33/min outside cascades, 0.77/min during domino cascades (0.46/min from domino cascade members), and 0.46/min during common-cause cascades (0.04/min from common-cause cascade members). Random disruptions continued during cascades, compounding their effects. Interrater reliability between the human expert and ChatGPT-4 was moderate (Cohen’s Kappa = 0.63, p < 10 −5 ). Conclusion All three disruption mechanisms—random, common-cause, and domino—were observed. Common-cause cascades stem from persistent environmental factors and tend to be low-rate but long-duration. Domino cascades are high-rate and short-duration. Random disruptions occur across all states, compounding overall disruption. The moderate agreement between human and AI detection suggests that automated tools may support early identification of error-prone states and enhance proactive surgical safety strategies, while the consideration of disruption cascades may help focus safety and efficiency improvements.